Update base for Update on "[ET-VK] Implement linear_q4gsw"

As title. Extend the quantized linear implementation to be able to handle 4-bit per group symmetrically quantized weights. This is in preparation to support using the int8 dot product extension to be able to handle dynamically quantized inputs.

Differential Revision: [D81800023](https://our.internmc.facebook.com/intern/diff/D81800023/)

[ghstack-poisoned]

Author: ssjia